Research Of Personalized Music Recommender System

When the users only have vague listening demands,a personalized recommender system can pick up some potential songs that the users might be interested in from tens of millions songs in the music library.Music recommendation algorithm contains three types:content-based recommendation algorithm,collaborative filtering recommendation algorithm and hybrid recommendation algorithm.Single recommendation algorithm cannot make full use of the huge amount of user information and item information in the system,and the mixing of different algorithms or rules can improve the accuracy of recommendation.Since music recommendation system has a much larger number of users than the number of songs,it can be significantly reduced by using collaborative filtering based on items when calculating the similarity of songs.Based on the collaborative filtering,the main research work of this paper is as follows:First,to calculate the similarity between items,item-based collaborative filtering need to make full use of the items that users prefer,since there are many behaviors that can reflect song preferences,this paper designs preference scores to get better song similarity accordingly.Second,because collaborative filtering only considers the user's behavior,and does not consider the correlation between the content of the songs,this paper integrates the interest label to carry out the algorithm's mixing to achieve better recommendation accuracy.Third,the deep neural network can fully learn the features of users,the features of songs,and the user's behavioral features of songs,so as to accurately predict the rate of song cutting.In this paper,we introduce a deep neural network to predict the rate of song cutting,and train a prediction model of cutting rate by using massive user listening data.The recommendation can be made more accurate by predicting and filtering the songs that are recommended by the hybrid algorithm.